Overconfidence Behavior and Dynamic Market Volatility - Core

Available online at www.sciencedirect.com

ScienceDirect Procedia Economics and Finance 13 (2014) 128 – 142

1st TSFS Finance Conference, TSFS 2013, 12-14 December 2013, Sousse, Tunisia

Overconfidence behavior and dynamic market volatility: evidence from international data Mouna Jlassia,*, Kamel Naouib and Walid Mansourc a b

Ecole Supérieure de Commerce, Université de Manouba, Tunisia Ecole Supérieure de Commerce,Université de Manouba, Tunisia c Islamic Economics Institute, King Abdulaziz University, KSA

Abstract

This paper examines the effect of overconfidence behaviour on dynamic market volatility in global financial markets. Using daily data from 27 countries spanning over 2000-2012, we find that the overconfidence is more pronounced for the advanced markets relatively to the emerging ones. With the exception of some Asian and Latin American markets overconfidence is present in both up and down markets. Evidence suggests that overconfidence is the main incentive that triggered and prolonged the global financial crisis in the US market and in other continents. Finding shows that overconfidence still exists even during the recession period, but at different levels.

© 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license © 2014 The Authors. Published by Elsevier B.V. (http://creativecommons.org/licenses/by-nc-nd/3.0/). Selection Selection and and peer-review peer-reviewunder underresponsibility responsibilityofofthe theTunisian TunisianSociety Societyfor forFinancial FinancialStudies Studies(TSFS). (TSFS) Keywords: Overconfidence; behavioral finance; global financial crisis; volatility; EGARCH.

1. Introduction The recent 2007-2009 global financial crisis is qualified as the most critical, dramatic crisis over the last decades (Shefrin and Statman, 2011). In addition, it has left a considerable long-term effect on market volatility and

* Corresponding author. Tel.: + 216 22 297 982 E-mail address:[email protected]

2212-5671 © 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/). Selection and peer-review under responsibility of the Tunisian Society for Financial Studies (TSFS) doi:10.1016/S2212-5671(14)00435-3

129

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142

was judged as the greatest recession since the Great Depression of the 1930s (Authers, 2010). A booming theoretical literature tackled this feature. Indeed, a bulk of modern empirical papers provided plausible explanations of the stylized facts of volatility and the financial crisis. Particularly, the behavioral prediction outlines that the persistence of financial instability in terms of excessive asymmetric volatility finds its roots in the change in our thoughts’ schema (Keynes, 1936). Overconfidence is regarded as the most prominent bias that relies on the core of volatile beliefs (Akerlof and Shiller, 2009). The behavioral literature asserts that it is one of the most robust behavioral findings (Shefrin, 2008) and a persistent dynamic phenomenon (Shiller, 1999). Overconfidence corresponds to individuals who are too confident and exaggerate in estimating their own competence and underestimate risk (Thaler, 2005). Such psychological pitfalls trigger an autocorrelation in investor’s errors beliefs about stocks’ intrinsic values (De Grauwe, 2012), which accelerates stocks mispricing and fuels market bubble. Shiller (2000) defines overconfidence as trading too much. Furthermore, several authors (e.g., Authers, 2010; De Grauwe, 2012; Abbes, 2012; Akerlof and Shiller, 2009) assert that overconfidence is the main incentive of investor’s beliefs change and presents the vital ingredient of fueling and riding bubbles. Despite the empirical evidences advanced on the magnitude impact of overconfidence on investor’s investment decisions among European and Americans markets, little findings have been documented on emerging ones (Chi, 2013). Especially, there is a limited exhaustive analysis of dynamic overconfidence effect on financial market evolution around the world during and after the financial crisis that erupted in 2007. This article contributes to the literature of behavioral finance in several ways. Indeed, at last five aspects are tackled. First, the originality of the paper stems from the investigation of the potential causes of clustered and asymmetric volatility witnessed before, during the global financial crisis, and during the recession period (i.e., postcrisis). Second, the dataset used is considerably large since it encompasses 27 financial markets categorized into four categories, namely: (i) advanced; (ii) Latin Americans; (iii) Asian; (iv) European & Mideast African markets. This large sample allows a deeper evaluation of the magnitude of US Subprime crisis on global financial markets behavior. Third, we scrutinize the role of US market as the main trigger of the global financial crisis. Fourth, we survey the role of overconfidence behavior in prolonging the Subprime crisis and leaving a considerable long-term effect on market conditional volatility. More particularly, we study the effect of overconfidence bias on the deep economic recession. Fifth, the empirical design focuses on the conditional volatility of stock prices based on a univariate EGARCH model by including a supplementary overconfidence component. The remaining sections are organized as follows. Section 2 expounds the benchmark model of overconfidence measure and its contribution to market conditional volatility. Section 3 presents the data collection and statistical analysis. Section 4 reports the empirical evidence. Section 5 examines the overconfidence variation under different market conditions. Section 6 summarizes. 2. Overconfidence estimation 2.1. Background on the estimated model Several studies in line with the dynamic model of Gervais and Odean (2001) assert that past realized success (i.e., high returns) stimulates investor’s overconfidence especially when such a success is confirming their private information. Thus, an overconfident investor becomes less risk-averse and trades excessively, which brings forth abnormal high trading volumes and impels stock prices to excessively increase. Indeed, Ko and Huang (2007) affirm that the presence of overconfident investors leads often to excessive volatility in stock prices. Deaves et al. (2010) maintain that such a behavior can be elucidated through the positive, significant causality relation according to which positive past returns affect trading volume. In order to distinguish the excessive trading component dictated by the investor’s overconfidence from the other factors affecting volatility, we implement Chuang and Lee’s (2006) methodology. We start by decomposing the trading volume into two components. The first is associated with investor’s overconfidence (OVER), whilst the second is not associated (NONOVER), as described in the following equations: ௣

௣

ܸ௧ ൌ ߙ ൅ σ௝ୀଵ ߚ௝ ‫ݎ‬௧ି௝ ൅ ߝ௧ ൌ ൣσ௝ୀଵ ߚ௝ ‫ݎ‬௧ି௝ ൧ ൅ ሾߙ ൅ ߝ௧ ሿ

(1)

130


(2)

ܸ௧ ൌ ܱܸ‫ܴܧ‬௧ ൅ ܱܱܸܰܰ‫ܴܧ‬௧

where ܸ௧ is the detrended market trading volume at time t (empirically estimated by the natural logarithm of number of shares traded in a day to the number of shares outstanding at the end of that day), ‫ݎ‬௧ି௝ is market return at time‫ ݐ‬െ ݆,ߚ௝ is a coefficient capturing the relationship between lagged stock return and actual trading volume, ߙ is a constant term, p measuresthe number of lags, and ߝ is an error term. Daniel, Hirshleifer, and Subrahmanyam (1998), Odean (1998), and Chuang and Lee (2006), among others, assert that the component ܱܸ‫ܴܧ‬௧ increases past stock returns and entails a high volume of transaction. However, the authors do not specify the number of lags in the relation between the stock return and the transaction volume. Thus, we resort to Akaike Information Criterion (AIC) and Schwarz Information Criterion (SC) in order to determine the ିଶ௅ ଶ௞ ିଶ௅ ௞௟௢௚௡ ൅ and ܵ‫ ܥ‬ൌ ൅ , number of lags (p). Formally, these two criteria are defined respectively as AIC = ௡ ௡ ௡ ௡ where k corresponds to the number of estimated variables, n corresponds to the number of observations, and ‫ܮ‬ represents the maximum likelihood function. Specifically, the determined number of lags in our sample is not identical for all countries but ranges from two (p = 2) to five lags (p = 5). 2.2. Overconfidence contribution to observed market conditional variance In order to capture the asymmetric character of abnormal high volatility, we use the univariate EGARCH model, as defined in Nelson (1991). The EGARCH (1,1) model captures successfully the asymmetric response in the conditional variance (Alexander, 2009; Lamoureux and Lastrapes, 1990; Chuang and Lee, 2006; Alexander, 2009; Abbes, 2012). It is presented as follows: ܴ௧ ൌ ߤ௧ ൅ ߟ௧ ߟ௧ ȁሺܸ௧ ǡ ߟ௧ିଵ ǡ ߟ௧ିଶ ǡ ǥ ǡ ܴ௧ିଵ ǡ ܴ௧ିଶ ǡ ǥ ሻ ‫ܦܧܩ ׽‬ሺͲǡ ݄௧ )

(3)

ȁߟ௧ିଵ ȁ ൅ ‫ߟܭ‬௧ିଵ ݄௧ ൌ ߱ ൅ ݂ଵ ቈ ቉ ൅ ݂ଶ ݄௧ିଶ ඥߟ௧ିଵ where ܴ௧ : the market return at time t. ߤ௧ : the mean of ܴ௧ conditional on past information. ߟ௧ : the residual of the mean at time t. ݄௧ : the conditional volatility at time t. ݂ଵ : the volatility effect on conditional volatility. ݂ଶ : the GARCH effect or the persistence in conditional volatility irrespective of any event happening in the market. ௙ ି௄ K : the asymmetric effect of the model. The Asymmetric Degree (AD) statistic is given by ቚ భ ቚ. ௙భ ା௄

߱ : a constant term.

The conditional volatility in the EGARCH model is modeled to capture the asymmetry effect of volatility. This asymmetry is estimated by the volatility parameter K. When ‫ ܭ‬is significantly different from zero, then response of volatility is asymmetric. If K< 0, then positive shocks (bad news) have larger reverberation on increased volatility than positive shocks (positive news). Thus negative stock return tends to increase more intensively volatility positive ones. Estimating the impact of overconfidence bias on conditional volatility consists in regressing the variable ܱܸ‫ܴܧ‬௧ on conditional variance. In line with Chuang and Lee (2006), we include both components from equation (1) into the conditional volatility model based on EGARCH (1,1), which is formally expressed as: ܴ௧ ൌ ߤ௧ ൅ ߟ௧ ߟ௧ ȁሺܸ௧ ǡ ߟ௧ିଵ ǡ ߟ௧ିଶ ǡ ǥ ǡ ܴ௧ିଵ ǡ ܴ௧ିଶ ǡ ǥ ሻ ‫ܦܧܩ ׽‬ሺͲǡ ݄௧ ) ȁఎ೟షభ ȁା௄ఎ೟షభ

݈݄݊௧ ൌ ߱ ൅ ݂ଵ ൤

ඥఎ೟షభ

൨ ൅ ݂ଶ ݈݄݊௧ିଶ ൅ ݂ଷ ܱܸ‫ܴܧ‬௧ ൅ ݂ସ ܱܱܸܰܰ‫ܴܧ‬௧

(4)


131

where ݂ଷ : corresponds to the overconfidence effect on conditional volatility. ݂ସ : corresponds to the effect of other factors besides overconfidence on conditional volatility. Research on volatility excess typically examines the link between trading volume and return volatility of stocks under the prediction that there is a increasing causal (lead–lag) relationship between the transaction volume and stocks return’s volatility (Darrat et al., 2007; Kurz, 2009). The positive and significant sign of the coefficient ݂ଷ implies that conditional variance should synchronically increase with the rise of transaction volume dictated by overconfidence behavior of the trader (Shalen, 1993; Kandel and Pearson, 1995; Karpoff, 1987). We accordingly expect ݂ଷ to be positive and significantly different from zero. Gervais and Odean (2001) examine the overconfidence in a multiperiod model and confirm a positive correlation relating past return to transaction volume. They assert that the realized market loss reduces slightly the investor’s overconfidence. Further, the authors outline that the overconfidence bias involves a different degree and this bias exhibits dynamic changes over time depending on past success or failure of the investor. If the overconfidence behavior provides a plausible explanation to clustered and asymmetric volatility phenomena, we then expect that ݂ଷ ൐ ݂ସ ൐ Ͳ. 3. Data collection and statistical analysis The dataset employed in this study is based on daily market price indices, daily transaction volumes of market indices from twenty seven countries. The sample covers: (i) Eleven advanced markets including United States (US), the United Kingdom (UK), New Zealand (NL), Switzerland (SZ), Japan (JP), Hong Kong (HK), Finland (FL), Germany (GR), Canada (CN), France (FR) and Australia (AU). The indices we use are respectively: S&P500 (US), FTSE100 Index (UK), ZSE50F (NL), SMI Index (SZ), NIKKEI 225 Index (JP), HSI (HK), OMX Helsinki 25 (FL), DAX (GR), S&P/TSX 60 (CN), CAC40 (FR), and AS51 Index (AU). (ii) Four Latin American markets including Argentina (AG), Brazil (BR), Chile (CL), Mexico (MX). The indices we use are respectively: MERVAL Index (AG), IBOV Index (BR), IPSA Index (CL), and MEXBOL Index (MC). (iii) Seven Asian markets including China (CN), Philippines (FP), India (ID), South Korea (KR) , Malaysia (ML), Thailand (TA) and Taiwan (TW). The indices we use are respectively: SHSZ300 Index (CN), PCOMP (FP), NIFTY Index (ID), KOSPI2 Index (KR), FBMKLCI Index (ML), SET Index (TA), and TWSE index (TW). (iv) Five emerging markets from Europe and Mideast Africa including Turkey (TR), South Africa (SF), Russia (RS), Kuwait (KW) and Egypt (EG). The indices we use are respectively: Istanbul XU100 Index (TR), FTSE/JSE Index (SF), RTS Index (RS), KWSEIDX Index (KW), and EGX 30 Index (EG). The dataset starts from January 5, 2000 to December 12, 2012, which constitutes 3344 observations for all countries, except for New Zealand, China, South Africa, Egypt, and Kuwait. Because of limited availability of data, the starting date is respectively as follows: New Zealand (March 7, 2003); China (April 12, 2000); South Africa (October 7, 2002); Egypt (February 3, 2004); and Kuwait (August 25, 2001). All the data are extracted from Bloomberg and DataStream International. Table1 and Table 2 report respectively the summary statistics of market index return and trading volume for the countries in our sample. According to Table 1, the descriptive statistical analysis of daily market return reveals that the average return in panel A is negative for all advanced markets, except for Finland and Australia that exhibit positive and very small values. However, the average trading volume for advanced markets in Table 2 shows a large volume with a positive average. This reveals that investors in developed markets tend to trade excessively and unprofitably. More particularly, the American stock market has the largest trading average (0.005) and the lowest average return (0.0002), which insinuates the presence of overconfidence behavior among American investors.

132

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142 Table 1. Summary statistics of global market return. Mean

Median

Max

Min

Std. Dev.

Skewness

Kurtosis

J-Bera (Prob)

Panel A: Advanced markets US

- 0.0002

0.0007

0.1095

-0.0946

0.0129

-0.281

13.710

11814.42 (0.000)

UK

- 0.0001

0.0005

0.0938

-0.0926

0.0123

-0.1328

11.372

7219.06 (0.000)

NL

0,0003

0.000

0.058

-0.049

0.007

-0.423

8.173

2823.3 (0.000)

SZ

-0.0001

0.0006

0.1078

-0.0810

0.0116

0.0064

10.916

6511.22 (0.000)

JP

-0.0001

0.0003

0.7136

-0.4054

0.0230

8.8353

424.196

HK

-0.0002

0.0005

0.1775

-0.167

0.0165

0.1223

20.186

3037.4 (0.000)

1826.32 (0.000)

FL

5.59E-05

0.0007

0.0885

-0.092

0.015

-0.197

7.173

1808 (0.000)

GR

- 0.0004

0.0009

0.1079

-0.0743

0.0146

0.047

9.031

3762.52 (0.000)

CN

-0.0002

0.0008

0.1563

-0.1731

0.0132

-0.943

29.764

3739.14 (0.000)

FR

-0.0001

0.0004

0.1059

-0.094

0.014

0.081

9.579

4481.01 (0.000)

AU

0.0001

0.0005

0.056

-0.087

0.011

-0.475

8.697

3462.75 (0.000)

Panel B: Latin American markets AN

0.0004

0.0008

0.1611

-0.129

0.020

-0.207

8.5965

4388.108 (0.000)

BR

0.0003

0.0009

0.1367

-0.1209

0.0185

-0.1627

7.1188

2378.489 (0.000)

CL

0.0003

0.0006

0.1180

-0.071

0.0101

0.0512

12.5211

12632.24 (0.000)

MX

0.0005

0.0009

0.1044

-0.0826

0.0139

0.0480

7.748

3143.135 (0.000)

Panel C: Asian markets CN

1.0002

1.0005

1.0945

0.9115

0.0169

-0.1165

6.3884

114.586 (0.000)

FP

-9.13E-05

-0.0001

0.7045

-0.4054

0.0225

9.5499

4.99

206.73 (0.000)

ID

-0.0002

0.0013

0.0796

-0.1305

0.0168

-0.697

8.307

298.0261 (0.000)

KR

-0.0001

0.0010

0.1153

-0.1273

0.0193

-0.551

7.2708

192.02 (0.000)

ML

2.08E-06

0.0002

0.6906

-0.7009

0.0265

-0.381

4.55

202.304 (0.000)

TA

-6.98E-05

-1.95E-05

0.710

-0.405

0.023

8.950

4.29

179.944 (0.000)

TW

-0.0001

0.0001

0.065

-0.0691

0.0163

-0.0938

4.780

317.170 (0.000)

Panel D: Europe & MidEast Africa markets TR

0.0005

0.001

0.121

-0.0901

0.0174

-0.185

5.9204

780.421 (0.000)

SF

-0.0050

0.0011

0.4133

-0.6931

0.0218

-10.602

542.85

262.6

(0.000)

RS

0.0002

0.0018

0.2020

-0.2119

0.0229

-0.3754

14. 79

119.59 (0.000)

KW

0.00012

0.0005

0.6899

-0.6931

0.0225

-0.2129

819.68

601.67 (0.000)

EG 0.001 0.003 2.0629 -0.6948 0.0573 18.445 816.75 597.06 (0.000) This table list market descriptive statistics of daily market return : mean, median, maximum, minimum , standard deviation (Std. Dev), skewness, kurtosis and Jarque-Bera (J-Bera) for eleven developed markets including United States (US), the United Kingdom (UK), New Zealand (NL), Switzerland (SZ), Japan (JP), Hong Kong (HK), Finland (FL), Germany (GR), Canada (CN), France (FR) and Australia (AU); four Latin American markest including Argentine (AG), Brazil (BR), Chile (CL), Mexico (MX); seven Asian markets including China (CN), Philippines (FP), India (ID), Korea (KR), Malaysia (ML), Thailand (TA) and Taiwan (TW) and five emerging market from Europe and Mideast Africa including Turkey (TR), South Africa(SF), Russia (RS), Kuwait (KW) and Egypt (EG). The data range is daily starting from January 5, 2000 to December 12, 2012.

The spread of return is large with a minimum value of (- 0.091) and a maximum value (0.106), which indicates the high volatility of U.S stock market. Indeed, we find that emerging markets exhibit higher average values of trading volume and lower average returns compared to those of advanced markets. This suggests that they have unusual volatile trading volume and subsequently a high negative return due to unexpected news. Moreover, by checking the minimum and maximum returns for emerging markets compared to developed ones it is noticeable that return variation is larger. This high variation may be due to the dynamic overconfidence bias. More particularly, the

133


difference between the maximum and minimum values is extremely larger for emerging markets, relatively to the developed ones. Similarly, the trading volume series seems to be more volatile than market return series, i.e., having higher standard deviation values. Table 2. Summary statistics for global market trading volume. Mean Median Max

Min

Std. Dev.

Skewness

Kurtosis

J-Bera (Prob)

Panel A: Advanced markets US

0.0005

-0.004

1.154

-1.549

0.211

-0.166

15.11

2832.35 (0.000)

UK

0.0001

0.002

2.185

-2.369

0.272

-0.350

1121

5320.43

NL

4.58E-05

0.0146

2.5495

-2.0669

0.405

-0.014

6.132

877.73

(0.000)

SZ

0.0001

0.001

1.535

-1.4512

0.3105

-0.128

4.601

489.6

(0.000) (0.000)

(0.000

JP

0.004

0.0003

1.026

-0.921

0.196

0.103

6.594

551.82

HK

0.001

-0.0046

2.4165

-1.931

0.3257

0.348

6.466

2348.16 (0.000)

FL

0.0003

-0.0019

2.297

-1.7680

0.3524

0.298

6.646

962.69

(0.000)

GR

0.0003

0.0029

1.652

-1.716

0.302

-0.089

11.065

453.44

(0.000)

CN

0.0002

0.0097

1.5286

-2.495

0.305

-0.281

8.958

2402.6

(0.000)

FR

-4.95E-05

0.0096

1.1532

-1.668

0.2688

-0.417

6.611

1026.6

(0.000)

AU

0.0003

-0.002

3.734

-3.595

0.3494

-0.063

7.33

1836.69 (0.000) 5636.92 (0.000)

Panel B: Latin American markets AG

0.0002

-0.0038

2.6630

-3.1603

0.5107

0.1411

9.3542

BR

-0.0002

-0.0044

3.0536

-2.7919

0.6824

0.0654

4.1528

187.56

CL

-0.0006

-0.008

1.8500

-2.2709

0.3314

0.0516

6.2593

1481.69 (0.000)

MX

5.89E-05

-0.0128

2.7447

-2.4525

0.4431

0.0910

4.9735

547.31

(0.000) (0.000)

Panel C: Asian markets CN

0.001

-0.011

7.735

-7.270

0.306

1.372

297.25

900.56

(0.000)

FP

0.0008

-0.017

3.159

-2.524

0.514

0.113

4.444

211.38

(0.000)

ID

0.0005

0.0075

2.738

-3.0076

0.239

-0.1725

28.156

580.121 (0.000)

KR

0.0005

-0.0056

2.0359

-1.629

0.2555

0.556

8.7202

3392.56 (0.000)

ML

0.0004

-0.0039

1.6310

-1.4245

0.2977

0.0438

4.2385

152.97

(0.000)

TA

0.0003

-0.0095

1.5807

-1.1242

0.3155

0.2407

3.6438

64.613

(0.000)

TW

-4.40E-05

-0.006

1.860

-1.729

0.231

0.261

8.895

4303.71 (0.000)

Panel D: Europe & MidEast Africa markets TR

0.001

-0.002

13.815

-7.172

0.437

12.817

11.43

222.93

(0.000)

SF

-0.0014

0.005

1.733

-2.0085

0.329

-0.347

5.697

999.19

(0.000)

RS

0.0015

-0.0100

5.4515

-6.773

0.8758

-0.1571

13.635

317.88

(0.000)

KW EG

-0.0075 0.0039

-0.0341 0.0001

2.178 6.2654

-2.137 -7.0497

0.324 0.4397

0.151 -2.0824

97.362 9.206

716.39 303.86

(0.000) (0.000)

This table lists market statistics of daily trading volume: mean, median, maximum, minimum, standard deviation (Std.dev), skewness, kurtosis and Jarque-Bera (J-Bera) for eleven developed markets including United States (US), the United Kingdom (UK), New Zealand (NL), Switzerland (SZ), Japan (JP), Hong Kong (HK), Finland (FL), Germany (GR), Canada (CN), France (FR) and Australia (AU); four Latin American markest including Argentine (AG), Brazil (BR), Chile (CL), Mexico (MX); seven Asian markets including China (CN), Philippines (FP), India (ID), Korea (KR), Malaysia (ML), Thailand (TA) and Taiwan (TW) and five emerging market from Europe and Mideast Africa including Turkey (TR), South Africa(SF), Russia (RS), Kuwait (KW) and Egypt (EG).The dataset range is daily starting from January 5, 2000 to December 12, 2012.

The normality test (probability of Jarque-Bera test) shows that both series of market return and trading volume are not normally distributed. Indeed, all probabilities reject the null hypothesis of normal distribution (Prob = 0); the skewness is different from zero and the kurtosis coefficient largely exceeds 3 for all countries. This result implies the presence of fat-tail in extreme return distributions and suggests the importance of asymmetric character

134


in stock volatility. The heavy-tailed return distributions seem to be higher for Asian and Europe & Mideast countries more particularly for Japan (kurtosis = 424.196) in advanced markets, Chile (kurtosis =12.521) in Latin markets, Philippines (kurtosis = 459.299) in Asian markets, and Kuwait (kurtosis = 819.68) in Europe & Mideast markets. The asymmetric trading behavior of investors in Table 2 shows that investors are prone to trade more aggressively in bull markets than in bear ones since the trading volume responds more intensively to bad news, which accentuates the asymmetric character of volatility in the American market and conveys it at the global financial market (Chuang and Lee, 2006; Glaser and Weber, 2009). 4. Empirical evidence 4.1. The reverberation of subprime crisis on market volatility The summary statistical analysis of market return and trading volume series provides insights that the behavior of indices in global market exhibits an extreme and abnormal high volatility. To investigate the asymmetric and clustering behavior of volatility, we use the univariate EGARCH (1,1) model. The estimation results are presented in Table 3. Table 3. Univariate EGARCH estimation of global market conditional volatility (2000-2012). Constant ࢌ૛ ࢌ૚ US -0.254* 0.105* 0.981* (-11.87) (9.58) (595.50) UK -0.227* 0.106* 0.984* (-11.72) (8.55) (578.56) SZ -0.314* 0.121* 0.976* (-11.95) (9.16) (444.53) NL -0.352* 0.134* 0.975* (-6.77) (8.86) (210.68) JP -0.251* 0.184* 0.987* (-34.10) (19.50) (914.90) HK -0.481* 0.242* 0.964* (-18.25) (33.75) (330.44) FL -0.132* 0.087* 0.992* (-18.42) (14.83) (1318.01) GR -0.275* 0.120* 0.979* (-10.92) (8.18) (471.94) CN -0.155* 0.100* 0.991* (-15.73) (13.28) (850.76) FR -0.226* 0.094* 0.982* (-11.25) (7.94) (556.13) AU -0.299* 0.133* 0.979* (-10.79) (10.81) (412.79) AG -0.363* 0.178* 0.970* (-13.52) (18.75) (355.59) 0.126* BR -0.327* 0.971* (9.65) (-9.02) (257.39) CL -0.637* 0.252* 0.952* (-10.29) (12.87) (166.38) MX -0.294* 0.141* 0.978* (-11.94) (10.63) (425.07) CN FP ID KR TW

K -0.131* (-17.02) -0.129* (-16.12) -0.142* (-17.02) -0.057* (-7.22) -0.073* (-10.36) -0.077* (-11.88) -0.059* (-11.59) -0.126* (-16.03) -0.073* (-10.54) -0.139* (-15.69) -0.118* (-15.19) -0.046* (-8.94) -0.085* (11.67) -0.075* (- 8.32) -0.098* (-12.80)

‫ۯ‬۲ 9.07 10.12 12.52 2.48 2.31 1.93 5.21 41 6.41 5.17 16.73 1.69 5.14 1.84 5.56

-0.189* (-7.424) -0.798* (-13.78)

-0.012*** (-2.44) 0.352* (34.68)

0.989* (399.17) 0.937* (150.03)

0.132* (10.526) -0.085* (-8.59)

1.2

-0.594* (-14.47) -0.386* (-11.30)

0.256* (19.33) 0.186* (15.45)

0.953* (229.28) 0.970* (270.45)

-0.109* (-13.10) -0.086* (-13.98)

2.48

-0.309* (-11.69)

0.163* (14.31)

0.978* (364.30)

-0.081* (-11.49)

2.97

1.63

2.72

135

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142 ML

-1.236* (-22.75)

0.969* (74.46)

0.916* (145.65)

0.538* (19.23)

0.28

TA

-3.447* (-23.65) -0.295* (-11.52) -0.468* (-11.77)

0.582* (29.35) 0.198* (18.14) 0.222* (14.36)

0.643* (40.48) 0.981* (342.48) 0.961* (229.17)

-0.021*** (-2.11) -0.050* (-8.14) -0.065* (-8.717)

1.07

-6.573* (-11.43) -0.125* (-16.49)

0.458* (11.90) 0.188* (21.91)

0.212** (3.04) 0.997* (890.76)

0.357* (9.72) -0.066* (-15.45)

0.12

-6.797* (-13.73)

-0.274* (-18.96)

0.174** (2.93)

-0.411* (-26.68)

0.20

TR RS SF EG KW

1.68 1.83

2.08

This table reports the regression results of conditional variance in global financial markets estimated by the univariate EGARCH (1,1) model: ȁఎ೟షభ ȁା௄ఎ೟షభ

݈݄݊௧ ൌ ߱ ൅ ݂ଵ ൤

ඥఎ೟షభ

൨ ൅ ݂ଶ ݈݄݊௧ିଵ , where K is the asymmetry coefficient,݂ଵ captures the impact of positive shocks and ݂ଶ captures the

GARCH effect or the persistence in conditional volatility irrespective of any event happening in the market. The terms ݂ଵ െ ‫ ܭ‬and ݂ଵ ൅ ‫ܭ‬ ௙ ି௄ measure respectively the sensitivity of conditional variance to negative and positive shocks. The Asymmetric Degree (AD) is give by ቚ భ ቚ. The ௙భ ା௄

figures in the parentheses are the t-statistics. Note:*, **, *** denote respectively the significance at 1%, 5% and 10% levels.

According to Table 3, the regression results show that ݂ଵ is positive and significant at the 1% significance level for all indices suggesting that the conditional volatility responds to innovation shocks. We note that the coefficient݂ଶ reflecting the GARCH effect is positive and highly significant at the 1% level, which implies that the shocks pushing variance away from its long run average will persist for a long time. Moreover, ݂ଶ is on average largely higher than 0.9 for almost all financial markets indicating a long memory effect in conditional variance. Accordingly, the volatility takes a long time to die out following the Subprime crisis in the global financial market. The term K reflecting the asymmetric effect is highly significant and negative. Accordingly, bad news has a larger impact on volatility than positive ones. The magnitude of the bad news compared to positive ones is estimated ௙ ି௄ by the variable AD = ቚ భ ቚ. AD is positive and different from zero for all financial markets. More specifically, the ௙భ ା௄

asymmetric effect is persistent for Australia (AD = 16.73), Brazil (AD = 5.56), Taiwan (AD = 2.97), and Egypt (AD=2.08). The large values of AD reveal that negative innovations are more destabilizing to volatility than positive ones for advanced and emerging markets. 4.2. Overconfidence effect on market volatility Once the asymmetric and clustered volatility phenomena are detected in the previous regressions, we move to investigate the overconfidence contribution to the observed stylized fact of volatility in global financial markets. The estimation results of overconfidence influence on global conditional variance are reported in Table 4. Table 4. Estimation of overconfidence behavior in global financial market (2000-2012). Constant ࢌ૛ ࢌ૜ ࢌ૚ US UK SZ FL GR FR

ࢌ૝

ࡷ

Ȥ²(p-value)

-0.144* (-9.52) -0.168* (-9.01) -0.221* (-9.86)

0.074* (7.82) 0.094* (7.47) 0.098* (7.07)

0.990* (799.15) 0.989* (633.97) 0.984* (550.72)

5.062* (4.99) 5.473* (7.64) 4.615* (6.81)

3.262* (6.34) 0.965* (12.54) 1.347* (18.72)

-0.042** (-2.67) -0.021 (-1.24) -0.044** (-2.62)

38.45 (0.00) 12.785 (0.00) 22.946 (0.00)

-0.130* (-7.71) -0.166* (-8.06) -0.158* (-8.96)

0.102* (8.43) 0.094* (7.01) 0.085* (7.18)

0.994* (725.44) 0.989* (567.35) 0.989* (622.84)

2.184* (5.54) 3.877* (5.93) 3.715* (6.19)

1.467* (41.64) 1.391* (18.09) 1.385* (20.28)

-0.037** (-3.20) -0.045* (-3.20) -0.054** (-3.86)

3.28 (0.06) 14.13 (0.00) 14.764 (0.00)

136

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142 AU

-0.229* (-8.84)

0.119* (10.12)

0.985* (444.60)

6.287* (6.19)

0.695* (9.41)

-0.083* (-8.82)

29.655 (0.00)

CN

-0.251* (11.05) -0.535* (-10.88) -0.402* (-6.59) -0.643* (-13.95) -0.196* (-9.19) -0.281* (-9.90)

0.129* (12.19) 0.306* (20.20) 0.135* (8.19) 0.299* (17.30) 0.149* (13.26) 0.144* (9.54)

0.984* (464.49) 0.965* (192.52) 0.970* (176.54) 0.953* (203.11) 0.989* (456.38) 0.980* (377.31)

15.483* (28.80) -4.964* (-15.28) 4.813* (3.65) 58.771* (11.52) -1.624* (-6.58) 0.990* (28.09)

0.727* (12.57) 0.573* (7.73) 0.583* (8.16) 1.878* (21.82) 1.272* (28.55) -0.333* (-1.04)

0.085* (9.15) -0.172* (-13.46) 0.006 (0.28) -0.181* (-16.62) -0.058* (-9.49) -0.113* (-13.76)

887.32 (0.00) 216.33 (0.00) 10.43 (0.00) 126.33 (0.00) 125.28 (0.00) 17.027 (0.00)

-0.391* (-9.15) -0.576* (-9.70)

0.155* (9.20) 0.232* (12.08)

0.966* (214.87) 0.957* (177.05)

1.001* (23.69) -0.139 (-1.20)

-2.607* (-7.37) 0.360* (8.25)

-0.133* (-12.73) -0.079* (-8.72)

92.942 (0.00) 19.784 (0.00)

CN

-0.533* (9.57)

0.231* (9.11)

0.959* (168.30)

-1.893* (-7.08)

2.717* (21.54)

0.078* (5.33)

284.34 (0.00)

ID

-0.596* (-14.09) -0.301* (-10.33)

0.251* (16.70) 0.165* (11.64)

0.952* (211.52) 0.978* (338.07)

10.757*** (1.86) -3.878* (-18.23)

-0.089* (-8.44) 1.022* (13.13)

-0.133* (-8.51) -0.069* (-8.46)

3.54 (0.059) 338.50 (0.00)

-0.866* (-20.50) -0.756* (-12.33) -0.371* (-10.78) -0.894* (-15.67)

0.232* (15.33) 0.347* (23.34) 0.199* (12.60) 0.507* (36.66)

0.922* (233.51) 0.943* (147.28) 0.974* (288.45) 0.946* (186.73)

-5.114* (-47.77) 12.953* (10.79) -1.185* (-9.77) 1.518* (27.41)

1.359* (21.63) 0.907* (25.03) 1.055* (23.47) -3.644* (-131.25)

0.074* (7.11) -0.001 (-0.11) -0.056* (-7.15) -0.114* (-11.32)

101.24 (0.00) 101.24 (0.00) 212.94 (0.00) 119.37 (0.00)

SF

-0.317* (-6.94)

0.124* (7.12)

0.975* (221.51)

2.185*** (1.97)

0.811* (13.61)

-0.071* (-3.77)

31.06 (0.00)

EG

-0.48* (-7.33) -0.599* (-13.92)

0.201* (10.02) 0.331* (20.87)

0.960* (137.41) 0.964* (240.94)

-2.442* (-11.04) 0.963* (25.77)

0.970* (20.43) -0.538* (-11.50)

0.009 (0.74) -0.069* (-9.63)

30.14 (0.00) 43.83 (0.00)

-0.488* (-18.92) -0.456* (-12.14)

0.291* (36.26) 0.231* (14.39)

0.965* (264.12) 0.964 (-0.66)

-0.473* (-6.07) 0.391* (240.94)

0.833* (126.82) -0.275 (15.71)

-0.077* (-8.04) -0.058* (-5.06)

25.67 (0.03) 2.72 (0.099)

JP NL HK AG MX BR CL

KR TA FP TW ML

KW TR RS

This table reports the regression results of conditional variance in global financial markets by incorporating overconfidence component estimated ି ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଷ ܱܸ‫ܴܧ‬௧ ൅ ݂ସ ܱܱܸܰܰ‫ܴܧ‬௧ . The Ȥ²-statistic with one degree of by the EGARCH (1, 1) model: ݄௧ ൌ ߱ ൅ ݂ଵ ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଶ ݄௧ିଵ ൅ ߠܵ௧ିଵ

freedom is used to test for the null hypothesis H0: ݂ଷ ൌ ݂ସ . The p-value is the probability of Chi-squared terms. The figures in the parentheses are t-statistics. Note:*, **, *** denote respectively the significance at 1%, 5% and 10% levels.

The regression results reported in Table 4 show that all variables remain highly significant. The terms ݂ଵ and ݂ଶ are larger, positive, and significant compared to those in Table 3. More particularly, the parameter of asymmetric effect, ș, is significant and negative which supports the validity of the model. From Table 4, we could see that the coefficient݂ଷ , corresponding to the overconfidence impact on volatility, is significant for all indices, except for Chile, which suggests the presence of overconfidence behavior among all global financial markets. The coefficient ݂ଷ is highly positive for advanced markets, except for Japan. Similarly, it is highly positive for emerging markets, except for Latin markets (Argentina), Asian markets (China, South Korea, Thailand, and Taiwan) and Europe & Mideast Africa markets (Egypt and Turkey) where the overconfidence component is significant but negative suggesting that investors in these countries are under-confident, i.e., pessimistic.

137


On balance, the dominance of the positive sign of the coefficient ݂ଷ in all markets implies that the conditional volatility increases synchronically with the transaction volume related to overconfidence component. Thus, we conclude that the overconfidence bias contributes to increasing clustered and asymmetric volatility across the world. Furthermore, we notice that ݂ଷ estimates are greater than those of ݂ସ , which means that the conditional volatility is more explained by overconfidence behavior than by any other factors (biases). We applied Wald Test in order to examine the null hypothesis H0: ݂ଷ ൌ ݂ସ . The test results show that the Chi-squared Ȥ²-test statistics are inferior to 5%. We therefore reject the null hypothesis. 5. Overconfidence and the global financial Subprime crisis Since our sample period contains the global financial crisis of 2007-2009, we re-fit Eq. (4) and split the whole sample period into three sub-periods. The first one precedes the subprime crisis. It is labeled the ‘tranquil’ period. It starts from January 4, 2000 until June 29, 2007. The second one corresponds to the global financial crisis, or ‘turmoil’ period, that first erupted in the United States on July 2, 2007 (Authers, 2010; Phillips and Yu, 2011) and ended in mid-2009 (NBER report, 2010) and later 2009 for the global financial markets (Authers, 2010; Hulbert, 2010; Abbes, 2012). The third one follows the 2007 global financial crisis (recovery period) lasts from December 2009 until December 2012. We refer to this sub-period as the ‘global recession’ period (Gore, 2010; Dhaoui and Quetat, 2012) or as Krugman (2011) labels “Lesser Depression”. Krugman (2011) considers the current world economic situation that should be a recovery period is a global recession, which is a prolongation of 20072009 financial crisis. We investigate the impact of overconfidence bias on the behavior of return and market conditional variance separately under different market conditions. The regression results before, during and after the Subprime crisis are reported in Tables 5, 6, and 7, respectively. 5.1 Overconfidence behavior under up market The overconfidence regression results presented in Table 5 are quite similar to the findings in Table 4. In regards to Table 5, the overconfidence correlation with conditional variance is positively and highly significant for developed markets except for Japan. More specifically, Hong Kong exhibits the highest degree of overconfidence (i.e., ݂ଷ = 106.77). Similarly, the Asian markets show high degrees of overconfidence, e.g., India (݂ଷ = 26.06) and Philippines (݂ଷ = 16.73), with the exception of Japan that exhibits the weakest overconfidence degree (݂ଷ = - 7.74). Such high overconfidence degrees could be interpreted in terms of aggressive trading volumes by Asian investors who are extremely overconfident and trade too much and too speculatively. The empirical findings are supported by previous studies (e.g., Chuang and Wang, 2005; Yates et al., 1998). Moreover, the Latin American markets, except for Argentina, exhibit higher level of overconfidence than Europe & Mideast Africa markets. Nonetheless, the finding of under-confidence (negative sign of ݂ଷ ) among China, Thailand, Taiwan, Malaysia, Kuwait, does not support the evidence advanced by Yates et al. (1997), Whitcomb et al. (1995), Chen et al. (2007), Abbes (2012), and Chuang and Susmel (2011), among others. Table 5.Overconfidence behavior and market volatility during tranquil period (2000-2007). Market Constant ࢌ૛ ࢌ૜ ࢌ૝ ࢌ૚ Advanced markets US -0.06** (-1.74) UK -0.15* (-7.06) SZ -0.23* (-7.53) FL -0.12* (-6.03) GR -0.20* (-6.24) FR -0.17* (-6.30) AU -0.35* (-6.32) CN -0.16* (-5.47) JP -0.34* (-5.66) NL -0.37** (-3.13) HK -0.95* (-11.55) Latin American markets

0.02** (1.50) 0.06* (4.43) 0.10* (4.98) 0.10* (6.64) 0.101* (5.06) 0.08* (5.09) 0.10* (6.51) 0.09* (6.47) 0.17* (7.57) 0.08* (3.59) 0.31* (12.40)

0.99* 0.98* 0.98* 0.99* 0.98* 0.98* 0.97* 0.99* 0.97* 0.96* 0.92*

AG MX

0.15* (10.50) 0.15* (6.16)

0.98* (358.1) 0.96* (178.44)

-0.21* (-7.80) -0.45* (-7.83)

(277.15) (554.64) (399.20) (638.40) (375.22) (426.48) (185.74) (370.10) (167.67) (88.45) (107.08)

3.68** 6.65* 4.56* 2.11* 3.18* 3.29* 9.81* 8.36* -7.74* 2.88* 106.77*

ࡷ

Ȥ²(p-value)

(1.37) (6.28) (4.80) (4.55) (3.47) (3.82) (5.37) (6.83) (-7.80) (2.31) (15.7)

2.72** (1.99) 0.87* (7.60) 1.32* (15.25) 1.40* (36.33) 1.20* (11.75) 1.23* (12.29) 0.68* (8.02) 0.78* (11.13) 0.86* (8.03) 0.62* (5.51) 1.87* (19.5)

-0.078**(-1.62) -0.01* (-0.58) -0.05** (-2.40) -0.01 (-1.27) -0.062* (-3.28) -0.05 (-3.03) -0.085* (-5.88) 0.02*** (1.58) -0.13* (-9.66) -0.06 (-1.41) -0.24* (-15.30)

22.88* 29.38* 11.42* 2.30** 4.57** 5.56* 24.56* 37.72* 76.92* 28 .97** 240.46*

(0.01) (0.00) (0.00) (0.01) (0.03) (0.01) (0.00) (0.00) (0.00) (0.05) (0.00)

-1.12* (-3.73) 1.05* (23.43)

1.27* (22.5) -0.03 (-0.06)

-0.04** (-5.49) -0.14* (-11.65)

57.54* 5.00*

(0.00) (0.02)

138

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142 BR -0.55* (-6.25) CL -0.76* (-5.89) Asian markets

0.09* (3.87) 0.22* (7.82)

CN -0.33* (-3.49) 0.18* ID -0.77* (-9.73) 0.24* KR -0.30* (-7.25) 0.16* TA -0.71* (-9.10) 0.22* FP -0.78* (-8.89) 0.41* TW -0.35* (-7.36) 0.19* ML -1.16* (-13.55) 0.66* Europe & Mideast Africa markets

(5.7) (11.39) (8.85) (9.90) (18.20) (9.39) (25.24)

0.94* 0.93*

(98.04) (76.57)

0.97* 0.93* 0.97* 0.93* 0.94* 0.97* 0.92*

(95.87) (106.34) (235.75) (126.76) (102.02) (118.08) (124.09)

1.02* 0.30*

(13.45) (6.07)

-1.85* (-4.40) 26.06***(2.86) -3.89* (-14.17) -5.30* (-42.68) 16.73* (11.23) -0.87* (-5.37) -3.85* (-115.6)

-4.37* (-5.91) 0.36* (1.02) 2.25* 1.01* 0.70* 1.47* 0.97* 0.72* 1.41*

(15.51) (10.21) (7.04) (18.54) (22.49) (12.15) (20.41)

-0.16* -0.06*

(-8.32) (-4.62)

0.09* (4.41) -0.19* (-8.49) -0.05* (-5.51) 0.11* (8.25) 0.03*** (1.67) -0.06* (-6.01) -0.16* (-12.78)

46.96* (0.00) 0.04*** (0.08) 83.20* 7.61* 176.55* 33.40* 112.46* 61.49* 72.24*

(0.00) (0 .00) (0.00) (0.00) (0.00) (0.00) (0.00)

KW -0.56* (-8.39) 0.30* (13.46) 0.96* (141.02) -0.22* (-3.10) 0.69* (13.64) -0.10* (-9.34) 103.11* (0.00) EG -0.21* (-3.49) 0.13* (6.02) 0.98* (154.48) -0.61 (-1.49) 0.79* (8.48) 0.02** (1.62) 98.8* (0.00) SF -0.37* (-4.25) 0.10* (4.47) 0.96* (109.33) 1.42*** (0.70) 0.84* (7.63) -0.06***(-1.99) 4.03** (0.05) TR -0.18* (-8.81) 0.13* (8.32) 0.99* (414.46) -0.31***(-2.73) 1.27* (23.01) -0.04* (-4.17) 11.01* (0.00) RS -0.60* (-10.93) 0.19* (12.46) 0.94* (158.46) -1.21* (-3.38) 0.09** (2.13) -0.05* (-4.77) 7.32** (0.03) This table reports the regression results of conditional variance in global financial markets by incorporating overconfidence component estimated ି ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଷ ܱܸ‫ܴܧ‬௧ ൅ ݂ସ ܱܱܸܰܰ‫ܴܧ‬௧ . The Ȥ² test statistic with one by the E-GARCH (1, 1) model: ݄௧ ൌ ߱ ൅ ݂ଵ ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଶ ݄௧ିଵ ൅ ߠܵ௧ିଵ degree of freedom is used to examine the null hypothesis H0:݂ଷ ൌ ݂ସ .The p-value is the probability of Chi-squared terms. The figures in the parentheses are t-statistics. Note:*, **, *** denote respectively the significance at 1%, 5% and 10% levels.

In regards to Table 5, the coefficient of the overconfidence variable (݂ଷ ) during the tranquil period (bull market or euphoria period) is not only positively significant but also higher than the overconfidence component estimated during the whole period in Table 4. This evidence supports the theoretical assumption that overconfidence is prone to be more conspicuous in bull markets in which overconfident investors continue to ignore systematically market warning signals and trade excessively. As a consequence, stock prices are over-estimated and stock volatility increases asymmetrically, which fuelled the housing bubble that erupted in July 2007 in the US. The persistence of conditional volatility during the tranquil period (݂ଶ in Table 5) remains positive and greater than that during the whole period (݂ଶ in Table 4). Likewise, the asymmetric effect detected by the coefficient K is highly significant and negatively greater than that during the whole period. This finding is attributed to the overconfidence effect. Being overconfident about their ability to select winning stocks, the investors overvalue the precision of their information and underestimate the likelihood of extreme events. Accordingly, they trade more excessively on winning stocks during the bull market relatively to the bear one, which impels the stock prices up beyond their intrinsic values and leads to a persistent shift from their Fundamentals values. Such a behavior triggers an abnormal increase in stock volatility (Chuang and Lee, 2006; Ben-David et al., 2010; Shefrin and Statman, 2011). The variable NONOVER is positively signed, except for Mexico and Brazil. 5.2 Overconfidence during the Subprime crisis A distinctive finding emerging from Table 6 is that the overconfidence component expected to be null or negative is significantly positive and higher for the American market (݂ଷ = 5.39) during the Subprime crisis relatively to the tranquil period (݂ଷ = 3.68). This highlights the presence of illusion of control bias due to overconfidence among American investors. In other terms, they behave as if they possess some control on a complicated situation (i.e., the Subprime crisis) believing that they may influence events and control the situation (Langer, 1975; Shefrin, 2010). As a consequence, they continue to trade excessively during the crisis period, which results in a continuous, asymmetrical increase of the clustered volatility and prolongs the crisis period. In regards to Table 6, it is not surprising to observe a higher estimate of ݂ଶ during the crisis period relatively to the tranquil period (Table 5) and during the whole period (Table 4). This implies that long term volatility is more volatile during the crisis period. The coefficient of ݂ଷ in Table 6 for advanced markets (except for Japan and Hong Kong) is positive and significant, but lower than that before the crisis period (Table 5). This entails that, even when an overconfident investor bears losses during crisis period, their overconfidence level decreases slightly but persists. This result is at odds with the empirical evidence reported by Abbes (2012). Furthermore, the overconfidence component ݂ଷ for Latin American markets, Asian markets (except Philippines) and for Europe & Mideast Africa markets is significantly negative. Specifically, Hong Kong market has an overconfidence term that shifts from ݂ଷ = 58.771 during the bull market to ݂ଷ = - 23.59 during the crisis. Such a

139


sudden shift is due to the fact that when an overconfident investor is confronted to the reality, they tend to be more frequently surprised and over-react (Moore and Healy, 2008; Shefrin, 2010). In fact, the over-precision in investors’ beliefs caused by overconfidence bias lead them to undervalue the frequency that they can be wrong. They generally end up very surprised by committing mistakes more often than they had believed and turn to be pessimistic (i.e., under-confident). Table 6. Estimation of overconfidence behavior and market volatility during subprime crisis (2007-2009). Market Constant ࢌ૛ ࢌ૜ ࢌ૝ ࡷ ࢌ૚

Ȥ²(p-value)

Advanced markets US

-0.18* (-4.86)

0.07* (2.97)

0.98* (264.54)

5.39**

(2.13)

3.16* (2.49)

-0.01

(-0.30)

3.12*

(0.00)

UK

-0.25* (-4.55)

0.08** (2.58)

0.97* (179.68)

5.34*

(3.88)

1.06* (5.70)

-0.002

(-0.05)

8.83*

(0.00)

SZ

-0.24** (-3.73)

0.12* (3.34)

0.98* (140.86)

6.039*

(4.21)

1.48 * (5.45)

0.043

(0.93)

9.81*

(0.00)

FL

-0.09** (-2.40)

0.07** (2.34)

0.99* (236.94)

2.186*** (2.07)

1.82* (8.98)

-0.057

(-1.53)

0.12*** (0.07)

GR

0.02** (3.87)

-0.01*(-7.87)

1.00*(2.79E+08)

2.766*** (2.35)

1.98* (11.24)

-0.063* (-2.49)

0.4***

FR

-0.10** (-2.91)

0.07* (2.56)

0.99* (254.99)

4.551** (3.02)

1.823* (9.06)

-0.007* (-0.18)

3.04*** (0.08)

AU

-0.48* (-4.33)

0.11** (2.31)

0.95*

2.23

(1.45)

0.90* (4.11)

-0.13*

0.717*

(81.73)

(-4.33)

(0.05) (0.03)

CN

-0.28* (-4.24)

0.12** (2.84)

0.97* (165.57)

11.69*

(5.82)

1.50* (7.75)

0.11*** (2.55)

25.93*

(0.00)

JP

-0.50* (-5.51)

0.31* (5.64)

0.96*

(99.85)

-2.47* (-3.96)

-0.87* (-6.32)

-0.20*

(-6.46)

5.39**

(0.02)

NL

-0.83* (-3.87)

0.22* (4.03)

0.93*

(45.37)

10.46*

0.60* (3.95)

0.15*** (2.52)

19.98*

(0.00)

HK

-0.31** (-3.34)

0.15***(2.78)

0.97*

(99.76)

-23.59 (-1.49)

-0.03

(-1.05)

2.61*

(0.01)

37.5*

(0.00)

(4.54)

1.89*

(7.92)

Latin American markets AR

0.03*

(5.65)

0.006 (0.72)

1.004* (856.91)

-3.42* (-4.81)

0.94* (7.83)

-0.08* (-11.42)

MX

-0.13* (-3.13)

0.08* (2.55)

0.99* (208.71)

0.21

(0.28)

1.19* (8.57)

-0.11*

(-4.78)

1.76*

(0.01)

BR

-0.41* (-5.12)

0.17* (3.94)

0.96* (104.59)

-0.70

(-0.77)

1.06* (10.74)

-0.14*

(-3.79)

3.92**

(0.04)

CL

-0.56* (-3.85)

0.25* (4.84)

0.95* (67.11)

-0.15

(-0.69)

0.41* (3.68)

-0.09*

(-3.82)

8.89**

(0.02)

Asian markets CN

-1.38* (-8.18)

0.10 (1.38)

0.83* (35.76)

-0.48

(-1.05)

3.29* (8.80)

-0.08*** (-2.23)

47.38*

(0.00)

ID

-0.81* (-6.53)

0.20* (5.01)

0.91** (64.55)

-7.33

(-0.62)

-0.31* (-8.89)

-0.10

(-2.33)

0.35*

(0.00)

(-5.18)

KR

-0.40* (-5.08)

0.10** (2.60)

0.96* (123.76)

-1.78** (-2.26)

2.15* (6.53)

-0.13*

21.49*

(0.00)

TA

-0.81* (-8.51)

0.17* (3.27)

0.92* (95.72)

-3.43* (-5.01)

1.08* (5.75)

0.05*** (1.72)

47.11*

(0.00)

FP

-0.75* (-4.53)

0.20* (4.44)

0.92* (50.25)

11.18** (3.01)

0.81* (5.68)

-0.004

7.82*

(0.00)

TW

-0.52* (-4.25)

0.14* (3.28)

0.94* (66.08)

-0.66***(-1.68) 1.50*

ML

-0.69* (-6.88)

0.09***(2.57)

0.93* (92.33)

-3.92** (-2.57)

(-0.11)

(6.19)

-0.04*** (-1.94)

41.44*

(0.00)

1.73* (12.62)

-0.04*** (-1.64)

13.87*

(0.00)

Europe & Mideast Africa markets KW

-0.35* (-4.10)

0.13* (2.95)

0.97*(111.59)

2.40

(0.97)

0.62* (4.79)

-0.11* (-1.78)

77.43* (0.00)

EP

-0.49* (-3.64)

0.20* (4.29)

0.95* (67.62)

-3.25* (-6.71)

0.31**(2.42)

-0.004 (-0.13)

68.55* (0.00)

SF

-1.02* (-4.31)

0.38* (4.51)

0.92* (42.56)

-1.16* (-4.13)

0.02 (0.30)

-0.16* (-3.38)

22.45* (0.00)

TR

-0.29* (-3.66)

0.04

(1.16)

0.96*(114.24)

0.48

(0.95)

1.74*(10.65)

-0.03***(-1.8)

9.62*

(0.00)

RS

-0.29* (-5.39)

0.17* (6.07)

0.97*(180.11)

-2.07**(-2.38)

0.1***(1.98)

-0.02 (-0.79)

9.72*

(0.00)

This table reports the regression results of conditional variance in global financial by incorporating overconfidence component estimated by the

140

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142 ି E-GARCH (1, 1) model: ݄௧ ൌ ‫ ݓ‬൅ ݂ଵ ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଶ ݄௧ିଵ ൅ ߠܵ௧ିଵ ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଷ ܱܸ‫ܴܧ‬௧ ൅ ݂ସ ܱܱܸܰܰ‫ܴܧ‬௧ . The Ȥ²-test statistic with one degree of freedom is used to examine the null hypothesis H0:݂ଷ ൌ ݂ସ .The p-value is the probability of Chi-squared terms. The figures in parentheses are tstatistics. Note:*, **, *** denote respectively the significance at 1%, 5% and 10% levels.

5.3 Overconfidence after the Subprime crisis The estimation results exposed in Table 7 shows that ݂ଶ remains positive and highly significant at 1% level. We also notice that the coefficient of asymmetric effect captured by ș remains unchanged for the emerging markets, but has slightly decreased for developed ones. For some developed markets (e.g., United Kingdom, Finland, Germany, Canada, and New Zealand), such an effect becomes positive. This fact shows that excessive and asymmetric volatility is present even in the no-news period. Indeed, the clustered, asymmetric market volatility is a persistent phenomenon in habitual periods (add reference here). The increasing positive value of ݂ଷ during the global recession period for developed markets (especially for the US), confirms that overconfidence is a persistent psychological pitfall. Previous studies do not confirm this result (e.g., Abbes, 2012). Table 7. Estimation of overconfidence behavior and market conditional volatility during recession period (2009-2012). Market Constant ࢌ૛ ࢌ૜ ࢌ૝ ࡷ ࢌ૚

Ȥ²(p-value)

Advanced markets US

-0.38* (-5.69)

0.11* (4.78)

0.96* (149.45)

8.192*

(2.26)

4.29** (2.39)

-0.063* (-1.29)

4.52**

(0.03)

UK

-0.13** (-2.96)

0.04***(1.68)

0.98* (237.30)

7.66*

(4.54)

0.97* (7.31)

0.01

(0.50)

15.78*

(0.00)

SZ

-0.17* (-4.11)

0.03***(1.90)

0.98* (259.42)

6.57*

(3.95)

1.31* (8.57)

-0.01

(-0.32)

9.89*

(0.00)

FL

-0.09* (-2.93)

0.04***(2.22)

0.99* (334.40)

6.02*

(4.26)

1.50* (9.41)

0.01

(0.54)

10.12*

(0.00)

GR

-0.05* (-2.06)

0.02***(2.41)

0.99* (449.84)

6.55*

(5.04)

1.46* (10.33)

0.02*

(0.97)

14.95*

(0.00)

FR

-0.50* (-5.35)

0.15* (4.47)

0.95*

3.36**

(2.46)

1.55* (10.46)

-0.14*

(-3.81)

1.77*

(0.00)

AU

-0.18** (-2.89)

0.04***(2.62)

0.98* (179.05)

5.07*** (1.90)

0.20

(1.02)

-0.12*

(-5.76)

3.19*** (0.07)

CN

-0.14* (-4.19)

0.04***(2.85)

0.98* (308.06)

25.93* (22.49)

0.62* (4.66)

0.13*

(5.93)

533.9*

1.66*** (0.09)

(97.07)

(0.00)

JP

-1.23* (-3.81)

0.17***(2.85)

0.87*

(26.27)

-1.44

(-0.55)

1.82* (6.48)

-0.16*

(-4.15)

NL

-1.68***(-2.95)

0.16* (3.42)

0.85*

(15.98)

9.53*** (2.14)

0.51* (3.25)

0.009

(0.14)

4.09**

HK

-0.17***(-2.37)

0.09***(2.87)

0.98* (147.02)

-43.01**(-1.62)

1.93* (9.99)

-0.03

(-1.12)

2.88*** (0.08)

(0.04)

Latin American markets AG

-0.19* (-2.77)

0.10* (4.13)

0.98* (126.15)

-1.47** (-2.06)

1.41* (14.04)

-0.06*

(-4.96)

16.74*

(0.00)

MX

-0.32* (-3.62)

0.16* (6.32)

0.97* (114.72)

-4.43* (-4.10)

0.66* (6.49)

-0.12*

(-6.37)

21.82*

(0.00)

BR

-0.14* (-2.56)

0.06* (2.97)

0.98* (157.84)

-1.54

(-1.03)

1.44* (11.68)

-0.07*

(-5.46)

3.97*** (0.04)

CL

-0.51* (-5.37)

0.18* (5.91)

0.96* (110.80)

-0.77** (-2.87)

0.26* (2.85)

-0.11*

(-7.23)

14.74*

(0.00)

Asian markets CN

-6.33* (-11.23)

0.13***(2.09)

0.31*

(4.95)

5.80*

(22.42)

3.62* (15.50)

-0.24*

(-4.59)

659.61* (0.00)

ID

-0.46* (-4.57)

0.08** (2.42)

0.95*

(93.19)

8.99*

(0.43)

1.16* (5.89)

-0.11*

(-2.62)

0.14*** (0.07)

KY

-0.62* (-4.84)

0.19* (4.39)

0.94*

(78.43)

-3.45** (-3.49)

2.10* (9.58)

-0.19*

(-6.03)

35.7*

TA

-1.42* (-8.32)

0.18* (3.55)

0.85*

(49.76)

-5.46* (-8.82)

1.18* (6.15)

-0.06*

(-1.52)

110.13* (0.00)

FP

-0.93* (-4.85)

0.22* (6.05)

0.91*

(47.51)

-3.83

(-0.73)

0.63* (6.34)

-0.15*

(-3.80)

0.74**

(0.03)

TW

-0.70* (-6.38)

0.13** (2.95)

0.93*

(85.34)

-1.37** (-2.55)

2.73* (9.85)

-0.13*

(-6.55)

42.28*

(0.00)

ML

-0.13***(-1.65)

0.09* (3.83)

0.99* (148.87)

-11.90* (-4.58)

1.64* (10.72)

0.02*

(1.01)

27.07*

(0.00) (0.04)

(0.00)

Europe & Mideast Africa markets SF

-0.37* (-3.62)

0.11* (3.16)

0.96*

(97.61)

0.35*

(0.11)

0.95* (6.66)

-0.12*

(-2.87)

0.45**

EP

-0.45* (-3.76)

0.16

(3.51)

0.91*

(32.88)

-12.46* (-6.28)

1.64* (6.77)

-0.01

(-0.34)

43.55*

(0.00)

KW

-1.12* (-5.23)

0.28* (7.99)

0.91*

(57.09)

-20.01*(-10.38)

1.03* (8.05)

-0.03

(-1.44)

111.7*

(0.00)

TR

-0.57* (-9.71)

0.15

0.94*

(70.21)

-0.99** (-2.46)

0.14* (5.66)

-0.18*

(-5.70)

7.86*

(0.00)

(4.47)

141

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142 RS

-0.54* (-5.38)

0.11* (3.67)

0.94*

(86.61)

0.26

(0.22)

0.51* (6.87)

-0.10*

(-3.61)

0.43*** (0.08)

This table reports the regression results of conditional variance in global financial by incorporating overconfidence component estimated by the ି ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଷ ܱܸ‫ܴܧ‬௧ ൅ ݂ସ ܱܱܸܰܰ‫ܴܧ‬௧ . The Ȥ²-test statistic with one degree of E-GARCH (1, 1) model: ݄௧ ൌ ߱ ൅ ݂ଵ ൫ߟ ଶ ௧ିଵ ൯ ൅ ݂ଶ ݄௧ିଵ ൅ ߠܵ௧ିଵ freedom is used to examine the null hypothesis H0:݂ଷ ൌ ݂ସ . The p-value is the probability of Chi-square terms. The figures in the parentheses are t-statistics. Note:*, **, *** denote respectively the significance at 1%, 5% and 10% levels.

On another side, the persistence of negative signs of ݂ଷ among emerging countries and some developed ones (e.g., Japan and Hong Kong) shows that under-confident investors became very pessimistic and tend to overweight the future volatility and under-estimate the market recovery. The under-confidence degree is prevalent especially among Latin American and Asian markets (except China and India). The behavioral explanation of the change in the overconfidence degree is due to the investors short memory and fear biases. Hence, during government regulation such biases are prone to be wax and wane in strength. It is for this reason that the regulatory authorities may fail and financial recession persists (Hirshleifer, 2008; Shefrin, 2010). Under this heading, the overconfidence may play a key role in influencing investors’ sentiment and delaying the global economic recovery. It is noticeable that the constant variable and the NONOVER parameter are almost statistically significant and positive for all countries. Their non-nullity and significance suggest that, besides the overconfidence biases, there exist other potential factors (e.g., psychological pitfalls such herding behavior, representativeness, and disposition fact) that can offer additional explanation to the asymmetric and excessive volatility phenomenon during and after the Subprime crisis. 6. Conclusion This paper studies investor’s overconfidence effect on dynamic market volatility. A sample of 27 countries is divided into four sub-groups: Advanced Latin American, Asian, Europe & Mideast Africa markets. Using a daily dataset from January 5, 2000 to December 12, 2012, we provide robust empirical evidence on the existence of overconfidence with different degrees among global financial markets, except for Chile. This finding shows that the investor reasons only over the short term and the decisions are often dictated by their psychological forces. Thus, a large part of excessive and asymmetric volatility in the global financial markets is explained by the overconfidence bias. The latter is a dynamic factor that engenders an aggressive asymmetric trading volume and increases stock prices fluctuation, especially during the global financial crisis 2007-2009. Further, the empirical design shows that, under different market conditions, overconfidence bias is a driving force of market disturbance. The contribution of this paper could be seen from various prisms. First, the overconfidence provides a plausible explanation of stocks’ clustered, asymmetric volatility and the financial crisis. Second, the shift in overconfidence degree is elucidated before, during and after the financial crisis. Third, the overconfidence contributes to explaining the long memory effect in volatility process. Fourth, the overconfidence plays a key role in the change of investor’s sentiment to turn to be too pessimistic (i.e., under-confident), which may let the regulatory reforms poorly performing and prolongs the economic recession. Future research could build on our findings in several ways. Indeed, a variety of interesting questions remain unanswered whose most important one is related to using intra-daily data. Indeed, high-frequency data could be more useful to confirm our finds in terms of persistency of overconfidence. References Abbes, M.B., 2012. Does overconfidence bias explain volatility during the global financial crisis? , Transition Studies Review 19:3, p.291 312. Akerlof, G.A., and Shiller, R.J., 2009. Animal spirits, Princeton, NJ: Princeton University Press Alexander, C., 2009. Practical financial econometrics. John Wiley & Sons, Ltd. Authers, J., 2010. The fearful rise of markets: global bubbles, synchronization meltdowns and How to prevent them in the future, United States of America: FT Press. Ben-David, I., Graham, J., Campbell, H., 2010. Managerial miscalibration, NBER Working Papers Series #16215. Chen, G., Kim, K.A., Nofsinger, J.R., Rui, O.M., 2007. Trading performance, disposition effect, overconfidence, representativeness bias, and experience of emerging market investors, Journal of Behavioral Decision Making 20:4, 425–451. Chi, M., H., 2013. Private information, overconfidence and intraday trading behavior: empirical study of the Taiwan stock market, Applied Financial Economics 23:4, 325-345. Chuang, W.I., K.L., Wang., 2005. Overconfident trading of Asian investors, Working Paper, Tung Hai University.

142

Mouna Jlassi et al. / Procedia Economics and Finance 13 (2014) 128 – 142 Chuang, W.I., Lee, B.S., 2006. An empirical evaluation of the overconfidence hypothesis, Journal of Banking and Finance 30:9, 2489–2515 Chuang, Wen-I., Susmel, R., 2011. Who is the more overconfident trader? Individual vs. institutional investors, Journal of Banking and Finance 35:7, 1626–1644. Daniel, K., D., Hirshleifer, A., Subrahmanyam., 1998. Investor psychology and security market under- and over-reactions, The Journal of Finance 53:6, 1839–1885. Darrat., Ali F., Maosen Zhong, Louis T.W., Cheng, 2007. Intraday volume and volatility relations with and without public news, Journal of Banking and Finance 31:9, 2711–2729. De Grauwe, P., 2012. Booms and busts in economic activity: A behavioral explanation”, Journal of Economic Behavior & Organization 83:3, 484–501. Deaves, R., Lüders E., Schröder M., 2010., The dynamics of overconfidence: Evidence from stock market forecasters, Journal of Economic Behavior & Organization 75:3, 402–412. Dhaoui, A., Quetat. I., 2012. Animal spirits and volatility of the trading volume in international financial Markets between 2002 and 2011, Working paper, Social Science Research Network, University of Sousse. Gervais, S., T., Odean., 2001. Learning to be overconfident, Review of Financial Studies 14:1, 1í27. Glaser, M., Weber, M., 2009, Which past returns affect trading volume? , Journal of Financial Markets 12:1, 1–31. Gore, C., 2010. The global recession of 2009 in a long-term development perspective, Journal of International Development 22:6, 714738. Hirshleifer, D., 2008. Psychological bias as a driver of financial regulation, European Financial Management 14:5, 856–874. Hulbert, M.,. 2010. It s dippy to fret about a double-dip recession, available at: http://online.barrons.com/article/SB50001424052970203983104575367471896032724.html?mod=go oglenews_barrons Kandel, E., N. D., Pearson, 1995. Differential interpretation of public signals and trade in speculative markets, Journal of Political Economy 103:4, 831-872. Karpoff, J., 1987. The relation between price changes and trading volume: a survey, Journal of Financial and Quantitative Analysis 22:1, 109126. Keynes, J.M., 1936. The General Theory of Employment Interest and Money, London: McMillan. Ko, K.J., Huang, Z., 2007. Arrogance can be a virtue: Overconfidence, information acquisition, and market efficiency, Journal of Financial Economics 84:2, 529-560. Krugman, P., 2011. The Lesser Depression, The New York Times, available at: http://www.nytimes.com/2011/07/22/opinion/22krugman.html Kurz, M ., 2009. Rational diverse beliefs and market volatility, In Thorsten Hens and Klaus Reiner Schenk- Hoppé, ed., 440-502, Handbook of Financial Markets: Dynamics and Evolution, Elsevier, North Holland. Langer, E. J., 1975. The illusion of control, Journal of Personality and Social Psychology 32:2, 311-328 Lamoureux, C.G., Lastrapes, W.D., 1990. Heteroskedasticity in stock return data: volume versus GARCH effects, Journal of Finance 45:1, 221229 Moore, D. A., and Healy, P. J., 2008. The trouble with overconfidence, Psychological Review 115:2, 502-517. US business cycle expansions and contractions : Annual report 2010, NBER, Sept,2010, available at: http://www.nber.org/cycles.html, Nelson, Daniel B., 1991. Conditional heteroscedasticity in asset returns: a new approach, Econometrica 59:2,347-70. Odean, T.,1998. Are investors reluctant to realize their losses?, Journal of Finance 53:5, 1775í1798. Phillips, P.C.B, and Yu, J., 2011, “Dating the timeline of financial bubbles during the subprime crisis”, Quantitative Economics 2:3, 455-491. Shalen, C., 1993. Volume, volatility and the dispersion of beliefs, Review of Financial Studies 6:2, 405-434. Shefrin, H., 2008. A Behavioral Approach to Asset Pricing, Elsevier: United States of America Oxford: Academic Press. Shefrin, H., 2010. Behavioralizing finance, United States of America: Now Publishers Inc. Shefrin, H., and Statman, M., 2011.Behavioral finance in the financial crisis: market efficiency, Minsky and Keynes, Working paper, Santa Clara University. Shiller, R.J., 2000. Irrational exuberance, Princeton, NJ: Princeton University Press. Shiller, R.J., 1999. Human behavior and the efficiency of the financial system. In: J. Taylor and M. Woodford (eds.), Handbook of Macroeconomics, Amsterdam Elsevier. Thaler, R. H., 2005, Advance in behavioral finance, Princeton, NJ: Princeton university press. Whitcomb, M.K., Curley, P.S., Benson, G.S., and Onkal, D., 1995. Probability judgment accuracy for general knowledge: cross-national differences and assessment methods, Journal of Behavioral Decision Making 8:1, 51-67. Yates, J. F., Lee, J.-W., Shinotsuka, H., Patalano, A. L., & Sieck, W. R.,1998. Cross-cultural variations in probability judgment accuracy: Beyond general knowledge overconfidence? , Organizational Behavior and Human Decision Processes 74:2, 89-117. Yates, J.F., Lee J.W., Bush J.G., 1997. General knowledge overconfidence: Cross-general knowledge overconfidence: Crossnational variations, response style, and "reality.” Organizational Behavior and Human Decision Processes 70:2, 87-94.